Testes baseados em correlação canônica para avaliar a consonância de painéis sensoriais

Abstract

A sensory panel is considered to be consonant when all referees point out a product in a similar way. Thus, the consonance between the referees that compose a sensory panel is one of the necessary characteristics for the reliability of sensory analysis, and can be measured by its level of unidimensionality. In literature there are tests for unidimensionality of a sensory panel, but its application is restricted to the analysis for one sensory attribute at a time. Thus, the objective of this study was the generalization of five tests proposed to evaluate the line of sensory panels, namely: the Fujikoshi test, the Monte Carlo test for unidimensionality, two parametric bootstrap tests and Monte Carlo test of eigenvalues. Such generalization is to use the canonical correlation matrix, so that it was possible to infer about the consonance of panels considering all attributes simultaneously. The performance evaluation of the proposed generalizations, in terms of power and type I error rate, was done through Monte Carlo simulation. The generalization of Monte Carlo test of eigenvalues test was applied to assess the panel’s consonance from an experiment conducted by Pereira (2005). The Monte Carlo test of eigenvalues performed equally to or higher than the other tests and, therefore, it is recommended for the analysis of multivariate unidimensionality of sensory panels.